HW60
状元
I am not allowed to view http://plecoforums.com/viewtopic.php?f=6&t=614 (flashcard simplified) by Gato. Can I have permission?
Mapping my Understanding of SRS to Pleco Flashcard Settings
- Thread starter Jim Kay
- Start date
I still think SRS can accommodate sporadic study provided that you set realistic goals for how many cards you want to learn, and provided that the system is intelligent enough to pare back your study pool if you can't keep up.
Fundamentally, most of what you guys are describing are configurations that would be *way* too complicated for the average user to attempt; we have to streamline things, and that means cutting out some features. "Weighted" imposes a whole lot of added complexity on the structure of sessions that would make it next to impossible to do more elaborate things with SRS, and both that and manual scoring force us to use confusing ideas like "scores" instead of nice clear "intervals."
The vast majority of our users want a flashcard system that will do all of the work for them - manage their studies in an intelligent way and adjust to them without requiring anything more than a minimal amount of configuration / input - and getting rid of a few features to achieve that seems like a worthwhile trade. Any usage scenarios that our revamped system doesn't accommodate can probably be added back, but when we add them back we would add them back as actual features that tell you what they are. (a "cram session" is a good idea, but it ought to be a "cram session" button, not a combination of a dozen settings that happens to give you something resembling a cram session)
Not a permissions thing, actually, it's part of the old Palm/WM forum that's no longer available.
HW60
状元
I think with "Weighted" you mean the Card selection system Frequency-adjusted, which in my opinion is one of the more easy settings of Pleco's flashcard system, as it requires no additional settings, and the only thing you have to remember is that low score cards are selected more often than high score cards.
The really difficult settings are the Tweak parameters of the Scoring system, because even if you should know how scores develop under the different settings, you still do not immediately understand what settings are the right ones for you.
Theoretically SRS is an ideal system: you increase the time between 2 reviews of a card with every correct answer, until in an ideal future you on average have to review (number of cards) * (points per day) / (maximum score) cards per day. Unfortunately, after a break this number of cards per day increases roughly by a factor of the missing review days. If you want to catch up without increasing your daily number of cards, there are 3 problems:
- Which due cards will you leave out?
- When will these cards cards be a member of the SRS again?
- How do you treat them in the meantime?
I do not know yet how these problems will be solved in the new system.
My last question: If you use SRS with a Score filter and review cards with a score between 1 and e.g. 200, aren't you close to Gato's system in http://plecoforums.com/threads/best-way-to-use-flashcards.3150/#post-25483?
The complexity there is internal - it takes a lot of code to support Frequency-adjusted tests and it forces us to do some things in terms of how flashcard sessions are structured internally that make it difficult to do other things we'd like to do. And I just don't think it's a very good system - any form of SRS will at least ensure some sort of individualized card tracking, while Frequency-adjusted is weighted-random but still random - a card that needs more review might have a greater *chance* of coming up more often, but it's not guaranteed to. Handling a behind-schedule situation by simply doubling the SRS interval of every card in your system would still provide better results than a switch to Weighted.
Not 100% sure here either - we'd probably make the first one at least customizable. But the basic idea is that cards that you don't have the time to study are simply no longer part of your study pool; they'll be the first cards to be added back once you have time for them, but we won't keep more cards in the pool than you can handle.
HW60
状元
I would very much appreciate if you could do something for the cards out of the pool in the sense of "maintaining some knowledge about cards that had been learned before" as described above.
There'd still be card filters, I think, so you could easily set up a filter that would include only otherwise-excluded cards. But if you've got the extra time to review more cards than those officially due, it's probably better if you just let the SRS system give you some more cards.
HW60
状元
Unfortunately I cannot explain what I mean. I try with an example: if I normally would like to review 300 cards per session and by some break I have 1500 officially due cards, I still would like to do 300, and review 30 important cards of the remaining 1200 to maintain some knowledge of those cards.
Yes, and I'm saying that Frequency-adjusted is a less-than-optimal way to do this because it's too random - a relatively new card really shouldn't be coming up at all in these extra tests (if you don't have time to review it on schedule then it's not worth bothering to try to learn it until you do) and a well-learned card doesn't really need to either since it can probably be reviewed a few weeks late without any ill effect.
So the best cards to study in that extra window should be cards that are in the middle - old enough that it's worth continuing to try to remember them, new enough that you can't simply put them off for two weeks. And even if your priorities are different, you ought to be able to set those priorities specifically rather than just trusting it all to a weighted random algorithm.
The way I am coming to see the Pleco flashcard implementation is that there are far too many 'adjusting settings' that either modify or override each other. The effect of a 'correct' or 'incorrect' scoring depends on all that PLUS the items historical state.
It's bad enough that the effects of one setting on the others is insufficiently documented, but the item's historical state cannot, as far as I'm aware, be displayed by an ordinary user. In this situation, the flashcard system becomes far to complex to be even reasonably predictable. Too unpredictable, in my view, to be usable.
The time it would take me to understand it--and it seems it's going to keep changing--would be better spent studying Chinese.
It's bad enough that the effects of one setting on the others is insufficiently documented, but the item's historical state cannot, as far as I'm aware, be displayed by an ordinary user. In this situation, the flashcard system becomes far to complex to be even reasonably predictable. Too unpredictable, in my view, to be usable.
The time it would take me to understand it--and it seems it's going to keep changing--would be better spent studying Chinese.
Thank you for underscoring my point earlier in this thread about the need to simplify it. (though I might point out that the "box" a card is in in your system is also a form of historical state that also determines the effect of a correct or incorrect score, and that since many other users prefer our current approach to card scores, supporting your boxes would mean either forcing that system on all of them or adding still more options to allow for both)
Sure they can - tap on the card in Organize and then tap on Statistics to see all of the data that we're maintaining on the card. Though I'm not quite sure which settings you're referring to here - most settings are quite self-contained, as long as you understand a few broad concepts like 'score' and 'easiness' pretty much all of the tricky settings are doing something with one of those.
As I said in my other post, I think it would probably be best if we just cancel / refund your purchase, since it seems like you're looking for a very different sort of flashcard system than what we offer. So please contact support and we'll be happy to help with that.
I don't think that's necessarily true. Spaced Repetition is complicated. Spaced Repetition wouldn't be practical without a computer. The manual system I mentioned above is just the way people used physical flashcards before software became available. The tricky part is the interface to allow people to select between the old physical flashcard ways and the complicated new Spaced Repetition method.
I'm not too optimistic that Spaced Repetition can be modified for people who study irregularly. I've seen many discussions about this problem on this forum over the years (some of them in the links above). People recognize that it's a major shortcoming of Spaced Repetition, but nobody has been able to describe a clear practical way to overcome this problem. Mike, you are as familiar with Spaced Repetition as anybody, but you don't have a solution yet, either.
If Pleco's next flashcard system upgrade keeps only Spaced Repetition, I hope there will be a way to upgrade the rest of Pleco without the flashcards. Otherwise, I probably won't be upgrading since the tight integration between the dictionary and the flashcard system is one of the biggest attractions about Pleco and on iOS, once you upgrade, it will be very difficult to downgrade back to the old version.
We'd keep Random around too - we're not going to force people to use SRS and nothing else. All I'm talking about getting rid of are a) Frequency-adjusted, because it's complicated to support and we don't see it as providing any benefits over SRS, and b) (probably) Fixed, because it seems like a bad idea pedagogically and the impression we get is that most of the people using it now would be just as happy with an option to casually flip through a list of cards in Organize without having half of their definitions cut off. Fixed is not too difficult to keep around and we may keep it initially simply because it'll lessen the blowback from getting rid of Frequency.
No, but I've got the beginnings of a solution now. All we're really trying to do is use your study time in the most efficient way possible; the mistake in SRS is to ask the user "how many cards do you want to learn" when we should be asking "how much time do you have to study today" - base everything off of that, since that's the thing over which we have the least control. Ideally, this would be flexible enough to also allow one to set a goal and be held to it (e.g. if you have a large word list and a limited amount of time to learn it and are willing to do what it takes to get there), but basically the idea is that we figure out how many cards you can learn / retain in the time you have to study and plan your reviews around that.
Would that apply if it keeps around Random too? Again, the only one we're firmly committed to getting rid of is Frequency-adjusted, since that's the one that causes the most added complexity, and we specifically *don't* want to get rid of Random since we certainly recognize the need to offer some sort of simpler alternative to SRS.
HW60
状元
After 2 days of testing a new setting I think I found a solution for me how to deal with SRS. Basically I had the idea after reading Gato's post about his system. I want to answer Pleco's question "how many time do you have to study today?" with "50 cards of my study pool", and I would like Pleco to set the Start Score Filter to Zero and adjust the End Score Filter to a value that lets me review close to 50 cards of my study pool. Actually you need not even change anything: just try 2 or 3 values to find an appropriate End Score filter with the desired result, and that is what I am doing now. I will not use Frequency-adjusted again nor any other Card selection system other than SRS!
The result of this setting is: no reduction of the desired number of cards to learn, temporarily (or permanently) review only lower score cards (if that were permanent it would probably better to reduce the study pool) and no pressure from SRS!
I do not know how far this is away from Gato's system - if there are tweak parameters that come close to his system, I could imagine that there is no big difference, because the main idea is to use a score filter.
I hope that that is only an option - I would like to choose the flashcard to review myself, and Pleco is not responsible for the result of my efforts
Cool, I'm very happy to hear that worked out. Since "Fixed" is still around (at the moment) and is seeming increasingly like it might be worth keeping if only for its score-based ordering option, you could also do a Fixed test in ascending order of score, with a 50 card limit - even more like Gato's system.
I hope that that is only an option - I would like to choose the flashcard to review myself, and Pleco is not responsible for the result of my efforts
Yes, this wouldn't be the only review system since we know people like to customize, but the default SRS behavior ought to make "how many cards do you have time to study right now" rather than "how many cards would you like to learn in general" its primary input.
Using the "Organize" feature to review cards would be very different between it doesn't keep track of what you get right or wrong and wouldn't be comparable to going through the flashcard tests.
I'd recommend keeping "fixed" because it replicate how physical flashcards work. "Random", like Spaced Repetition, is only possible with software and so may be similar enough to Spaced Repetition to be discard for the sake of simplicity. Then you would be left with a "fixed" order system that resembles physical cards, and a Spaced Repetition system that's very different from physical cards and uses a complicated algorithm to calculate the card order. That might be a good compromise between simplicity and flexibility.
That sounds like a combination of "fixed" selection and "automatic" (i.e. Spaced Repetition) scoring. Does the Spaced Repetition scoring makes it substantially different from the "manual" scoring system (moving up a level after getting 2 or 3 times correct in a row) that I used?
True, if people are using it for actual testing rather than just a "review only" scan through cards then that wouldn't work.
"Random" is very much achievable with paper cards, simply by shuffling them (as many people do anyway), so I think we'd want to keep that too. (and actually a "Fixed" system sorting by score seems like it would not be achievable with paper cards, at least not without great difficulty)
"Fixed" is similar to paper cards in multiple stacks. You would move a card to a new stack (higher level) once you get it 2 or 3 right in a row.
I suppose you could shuffle the paper cards at the beginning of each round of study to get the "random" effect, but you would have to shuffle at least once per review of the deck.
I suppose you could shuffle the paper cards at the beginning of each round of study to get the "random" effect, but you would have to shuffle at least once per review of the deck.
HW60
状元
There are actually 4 Card selection systems in Pleco, and Repetition-spaced is one of them - it is not a Scoring system. Card selection is always very simple, and so is Repetition-spaced: according to the score of a card and the "points per day" (standard is 100) the next review of a card comes after score/(points per day) days. With a card score of 1200 and points per day = 100 the next review comes after 1200/100=12 days.
One of the problems that should be cured with the next update is not the Card selection system Repetition-spaced, but the lots of details of the Scoring system "Automatic" (and besides there is also an Old Automatic and a Manual scoring system ...). Apart from exponentially and logarithmically scaling for early and late review, there are the Tweak parameters with lots of parameters which I guess less than 1% of the Pleco users ever understood. A long time ago I posted an Excel sheet to get some feeling for the development of the score under the various easiness factors and cushion unlearned - less functions would surely be more for the average user.
To answer your question: I think there should be some tweak parameters to adjust the Automatic scoring system so that the scores develop similar to your Manual scoring system - if not, you just choose Card selection Repetition-spaced combined with Manual scoring system to get the same scores. Then with Fixed Card selection the selection of cards is the same as you have it every day, with Repetition-spaced it is random. But the important difference between the Fixed Card selection system and the Repetition-spaced Card selection system is the amount of cards to be reviewed: Supposed you have a score filter from 0 to 500, then with a Fixed Card selection system you will have to review all cards with a score within these limits. With the Repetition-spaced Card selection system you also review cards with a score between 0 and 500, but only the due cards, and a card that got a score of 500 one day before will be due for review 5 days later. After a review break of 5 days or more the amount of cards to be reviewed should be the same in Fixed and Repetition-spaced, because then all cards with a score of 500 or less are due.
I probably have not yet understood the reasons for a fixed card selection system: if you have 10,000 paper flashcards, they are alltogether about 1 meter high, and I think the fixed handling of these cards would be impossible for me. Using Pleco with Repetition-spaced Card selection AND DAILY REPETITION OF THE DUE CARDS, you have a fair chance to review them all when they are due, but the real problem of Repetition-spaced is a different one, as I tried to describe above: when you have no time for the daily repetition.